This is my thesis project and I opened the research study settings using Lab.js. In this research, I designed a 2x2 mixed-subject experiment to investigate how the presence of ChatGPT label (between-subject factor: with vs. without) and the provision of explanations (within-subject factor: with vs. without) affect trust in AI over time. Participants engage in a complex decision-making task where they estimate jail time for 16 criminal law cases with AI advice. Hereby, behavioral trust could be estimated. Participants would also self-report their trust towards AI after each 4 cases. In this way we could also calculate their self-reported trust. These trust estimations were further utilized in data analysis to study the relations between trust in AI and various factors.
这里是我的毕业论文项目，开放给大家使用lab.js部署的研究设置。在本研究中，我采用了2 x 2混合主体实验设计，其中组间因素为ChatGPT与其他AI的对比，组内因素为是否提供解释。实验过程中，我要求参与者基于AI的建议来估算16个刑事案件的监禁时间，并且在参与者估算每4个案件后他们被要求自我报告对AI的信任度。基于这种方式我们可以计算出行为信任度和自我报告信任度。这些信任度此后被用于数据分析，以研究参与者对AI的信任度与各种因素之间的关系。
Both behavioral and self-reported trust are measured. Here I used Weight on Advice (WoA) as a measurement for behavioral trust. Weight on Advice = (2nd Estmate - 1st Estimate)/(AI Advice - 1st Estimate).
信任度则通过行为观察（WoA trust，weight of advice）和自我报告（Self-reported trust）两种指标来衡量。WoA = （第二次估算值 - 第一次估算值） / （AI预估值 - 第一次估算值）。
I used lab.js as the platform to design my experiment. It's a simple, graphical tool to help you build studies for the web and the laboratory. Several javascript scripts should be added to customize the settings.
lab.js是一款简单的图形化工具以帮助我们设计线上实验，我在这里使用它作为实验设计的平台。为了实现个性化设置，我们还需要写入一些Javascript脚本。
If you are interested in the research result, please contact me: [email protected].
如果你对本研究结果感兴趣，请联系我：[email protected]。